Watching the small99 explanation video, the calculator obviously optimises for simplicity, and deoptimises for accuracy. The model assumes a lot of averages to do that, and rightly talks about the tool as a conversation starter rather than an accurate estimate.
I still think we need it clear when an estimate is an estimate though, and without open source models, it's much harder to judge which parts of operations are more or less "average" than others, and what needs revising.